Robotics research vision would pair AI agents with real-world evaluation

A participant in the effort describes a vision for using frontier AI agents to handle much of the implementation work in robotics research, alongside robotics expertise and infrastructure for real-world evaluation at scale. They say reproducing results can require costly compute, dataset curation, annotations and hardware, even when code is public.